Unverified50% confidenceFactExact time
过度对齐（Over-alignment）指模型安全护栏设置过于敏感，在完全合法的使用场景中也拒绝提供帮助的反向风险
1
Sources
50%
Confidence
Long-term
Relevance
8/16/2026
First Seen
Sources
Related Claims
VerifiedExcessive safety fine-tuning can cause models to over-refuse, undermining their practical utility.79% similarUnverifiedOver-refusal strategies reduce risk but harm model utility, especially in scenarios such as professional development and security research that require touching sensitive boundaries.70% similarUnverified安全研究者将系统行动能力远超其安全判断能力的潜在风险称为能力越界（Capability Overhang）67% similarUnverifiedExcessively frequent human intervention eliminates the efficiency advantages of automation, while an overly lenient approval scope fails to effectively prevent and control risks — this is the trade-off in HITL granularity design66% similarUnverified过度授权（如在只需生成文档时同时开放命令执行权限）会因提示词注入攻击或模型幻觉引入破坏性操作的安全风险65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/757385API
curl https://kongchang.com/api/v1/knowledge/claims/757385MCP
get_claim(id=757385)